While the Occupy Wall Street (OWS) movement has been gaining momentum, growing in terms of visibility, media coverage and sheer numbers of participants, it has had a difficult time “occupying” the Twitter trending topics (TTs) list. #OccupyWallStreet, the movement’s dominant hashtag, has never once hit the New York TTs list. Similarly, #OccupyBoston has trended all across the world, but never in Boston, which only saw the phrases ‘Dewey Sq’ and ‘Dewey Square’ trend.
Some point the blame at Twitter for censoring content, yet what seems to be happening is purely algorithmic. There’s often more than meets the eye when it comes to algorithmically generated TTs. In this post we dissect some of the dynamics at play, looking at all OWS related terms that have trended on Twitter since the start of the movement, their volume of appearance in tweets, and the times and locations they’ve trended. As in our previous studies on networked audiences and virality , by understanding the networked environment that enables information to spread, we gain valuable insight on why certain topics become more visible and how locations around the world affect each other in this game to maximize attention.
What’s in a Trend?
Whether you’re an activist, event organizer or even a marketer, TTs bring tremendous visibility to the message that you’re sending out. On Twitter, TTs have become somewhat of a status symbol, a signal of success of a person, conference or newsworthy event. They are controlled by an algorithm that publishes a constantly updating stream of phrases or hashtags for locations chosen by Twitter.
The precise algorithm for determining trends is private, but the basic thrust is that it’s not about volume, or else Justin Bieber would be forever trending. The algorithm adapts over time, based on the changing velocity of the usage of the given term in tweets. if we see a systematic rise in volume, but no clear spike, it is possible that the topic will never trend, as the algorithm takes into account historical appearances of a trend. The implications are clear:
- The longer a term stays in the trending topic list, the higher velocity required to keep it there.
- It is much easier for a term never seen before to become a Twitter trend, and finally
- It is extremely important to understand what else is happening in the region or network (if Kim Kardashian’s show is airing, you can forget about trending!).
Over the past four weeks, I’ve been tracking the usage of Twitter amongst the OWS movement, drawn to understand how and why OWS related phrases and hashtags trend in certain regions but not in others. For example, #OccupyWallStreet is the most commonly used hashtag since the start of the movement, yet the term trended in Vancouver, Portland, Italy and San Francisco. Contrary to intuition, It has never appeared as a trending topic in New York, where most of the action took place.
Another example is the term ‘Occupy Boston’, which trended in numerous cities across the US, but never in Boston. By analyzing the ebb and flow of trending topics along with tweets mentioning those topics, we draw conclusions about #OccupyWallStreet, ‘Occupy Boston’, and literally see how the movement is gaining momentum, visibility, and spreading across the world.
When such prominent hashtags or events do not make it to the trending topics list, it is easy to accuse Twitter of censorship. George Reese (@GeorgeReese) calls it in the early days of the protest:
Interesting. #takewallstreet is trending. #occupywallstreet is not. Pretty strong evidence of censoring.
In an email to Adrianne Jeffries at BetaBeat, Sean Garrett, head of commmunications at Twitter, wrote that “[Trending Topics] are the most ‘breaking’ and reward discussions that are new to Twitter. We are not blocking terms related to #occupywallstreet in any way, shape or form.” Twitter has repeatedly insisted that there’s no censorship whatsoever happening.
What we’re seeing is an outcome of a purely algorithmic mechanism, with its built in biases, hence not always intuitive or logical. The algorithm is affected by volume, changes over time, and is customized for every one of the 111 geographic regions chosen by Twitter. Lets take a look at the data.
#OccupyWallStreet – A closer look
For this analysis, I chose two of the many signals we analyze at SocialFlow:
- All tweets mentioning OWS related terms (such as: ‘#OccupyWallStreet’, ‘#OWS’, ‘Occupy Boston’, etc…) since September 25th.
- All OWS related Twitter trending topics (same set of terms as #1).
My goal was to understand the dynamics at play – how the geographic regions differed, and why certain trending terms appeared in some but not in other regions.
Competing for Attention
The plot below shows the number of tweets per hour that included one of six chosen hashtags since September 15th. In blue, we see the daily cycles of Tweets mentioning #OccupyWallStreet consistently appearing since the initial Wall Street occupation in mid September. Comparatively, the other hashtags hit the Twitter trending topics list at some point in time, had a clear spike along with a fast decline.
When analyzing why a topic did or did not make it to the trending topics list, we must always look at what else was happening at that time; what people chose to give their attention to. The plot above shows the sheer differences in velocity between the consistent, slowly accelerating #OccupyWallStreet and the other hashtags that draw more than 4 times the participation at their peak. When we look at sheer numbers of participants in hashtags like #WhatYouShouldKnowAboutMe or #ICantRespectYouIf both of which trended in NYC, it is clear that they’re extremely difficult to compete with. When analyzing why a topic did or did not trend, it is important to understand what its coming up against. An intuitive example: it is much harder to make a topic trend during the day in major US cities compared to night.
If we look at the dispersion plot of OWS related trending terms split by geographic region below, we see some interesting observations. First of all, the very first city in which #OccupyWallStreet trended is Madrid. Perhaps a sign of the Spaniard’s elevated interest in this story is due to the highly publicized protests that took place across Spain over the course of the summer. Other regions where Twitter users are drawn to the story include Italy which is on the cusp of economic collapse, Germany, Portland, Washington, Chicago, San Francisco and Boston. The following day the story peaks the interest of Twitter users in Vancouver most likely due to the publicity around Vancouver-based activists Adbusters affiliation with organizing the protest in Wall Street.
There’s a substantial gap between September 23rd and October 2nd, where no OWS related term made it into any of the trending topic list. We see a consistently rising volume of tweets, yet not enough velocity to make it into a trends list in any of the geographic locations. Additionally, except for a few minutes where ‘Occupy New York’ was trending in New York on September 26th, we see no appearance of OWS in New York trending topics. Another interesting data point is the prominence of OWS in the United Arab Emirates trending topics list between October 3th and 5th. Perhaps the arrest of 700 protesters in New York resonated with Twitter users in the UAE, drawing an unusually large number of reactions, keeping the topic trending for a period of two days.
While the OWS movement is far from organized, there’s much to be lost from not using the same hashtag or term. OWS members have been criticized for using multiple hashtags such as #ows, #OccupyWallStreet and #TakeBackWallStreet instead of sending out a clear message to just use one. The joint volume might not have sustained the trend, but would have definitely given it more visibility at the start.
The graph below shows the numerous OWS related trending topics, and when they trended. For example, #occupymcr (Manchester) was used before October 3rd, yet reached the Twitter trending topics list only on that day, and never again afterwards.
Keeping it Fresh
In contrast to the point above, if one seeks to keep a topic trending, it is important to change it every few days. As seen above, a gradual rise in velocity only makes it harder and harder for the term to trend. Take for example #OccupyWallStreet which trended in many cities across the world during the first days of the protests. From September 23rd, even with a consistent growth in user adoption, the rate of growth was not enough for the hashtag to trend again. And it has not ever since.
The key here is to take a hybrid approach:
- Consolidation – try to get all participants to use the same hashtag.
- Keep it fresh – if you’re interested in maintaining traction you must alternate hashtags every few days.
There’s nothing like a Police raid and hundreds of arrests to push a story’s visibility. Alexis Madrigal eloquently describes how he followed the turn of events through the Twitter hashtag #OccupyBoston – “Literally thousands of tweets tagged with #occupyboston were going into the ether last night…”. Yet the stream of posts about the Boston police raid was still dwarfed by the number of reactions posted about the airing of part 2 of “Kim’s Fairytale Wedding: A Kardashian Event” on E!. The #KimKWedding hashtag saw a staggering 5x the volume of tweets compared to those mentioning ‘Occupy Boston’!
Looking at the data we notice four different terms that trended on October 11th in reaction to the events in Boston: ‘Dewey Square’, ‘Dewey Sq’, ‘Occupy Boston’ and ‘#OccupyBoston’. The dispersion plot below highlights the time of day when each term was trending regardless of location. #OccupyBoston was clearly the most dominant trend in terms of volume, yet ‘Dewey Sq’ was the first related trend to appear, followed by the longer version – ‘Dewey Square’.
When we filter the data down to only trending terms in the Boston region, niether ‘#OccupyBoston’ nor ‘Occupy Boston’ make it to the list:
This is somewhat odd. It makes sense that Boston folks tweeted about the events happening in their city, many of them must have used the #OccupyBoston hashtag. Yet the only terms that reached the trending topics list for Boston on Oct. 11th were ‘Dewey Square’ and ‘Dewey Sq’. When I compare the total volume of tweets per minute during peak trafic that day, I see a 25 times more tweets that include the term ‘Occupy Boston’ compared to those that include ‘Dewey Sq’. Yet ‘Dewey Sq’ was trending most of the day in Boston, and ‘Occupy Boston’ never…!
Perhaps there were simply not enough people tweeting in the Boston region? Other trending topics that day in Boston include #ItsAShame, #MyFavoriteSongsEver, #ThingsPeopleShouldNotDo… These might have been too large and too new to compete with. Additionally, maybe there aren’t enough “advanced” Twitter users in Boston who’d be using the hashtag, while many would repost media or publish descriptive tweets mentioning ‘Dewey Square’, where the events took place.
I also wonder if the TTs algorithm takes into account TTs in other locations. For example if #OccupyBoston trends across cities in the US, does it make it harder for it to also trend in Boston?
Above is a dispersion plot of all locations where ‘#OccupyBoston’ or ‘Occupy Boston’ trended on October 11th. Portland was the first city to pick up the topic, while Austin also sustained it for a fair bit of the day. In the peak hours (2-6pm), the phrase trended across United States, during which the topic was one of the top 10 terms gaining velocity and attention across all content produced on Twitter by users in the United States.
In this day and age, where the slightest algorithmic tweak can shift the scale of attention given to one topic over another, how transparent should we expect services like Twitter to be with regards to the way they calculate, adapt and manage what bubbles up to the top? At last month’s NYC Strata Conference, Tim O’reilly discussed the importance of checks and balances, referring to Algorithmic Regulation as a way to make sure power is not abused. At this point, I’m highly doubtful that regulation is the solution, but it certainly would be an interesting debate to be had.
It is important to remember Eli Pariser’s argument from The Filter Bubble: just because you see content of certain type, doesn’t mean that everyone else does! Perhaps its time that companies exposed information about their ranking algorithms, striking a fine balance between exposing just enough information to enable users to make smarter decisions. Or perhaps we should be thinking more deeply about the relationship between editorialized and algorithmically generated content? Techmeme and Mediagazer use a hybrid approach. What would it mean for TTs to be part algorithmic part editorialized?
Questions? Feedback? Feel free to ping me on Twitter – @gilgul